Anthropic announced on Tuesday (6) an expansion of Cyber Verification Program (CVP), which allows verified cybersecurity professionals to use its AI models with fewer restrictions in defensive work and authorized testing.
The new structure divides the program into three levels and incorporates Project Glasswing, an initiative focused on critical software security. Participants will be able to access Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, as well as future models.
The Defense Access level covers activities such as incident response, malware analysis, reverse engineering and vulnerability identification. Companies, universities, governments, critical infrastructure operators and researchers can request access.
The Red Team Access adds penetration testing and authorized operations against systems of participating organizations. Activities capable of causing major damage or disruptions, such as ransomware deployment, will remain blocked.
Meanwhile, Specialized Access will have fewer restrictions and will be aimed at organizations that work with sensitive systems, including power grids, telecommunications, banking infrastructure and government networks. Approval will require a more in-depth review.
Glasswing found at least 129,000 vulnerabilities
The expansion comes after results obtained with Project Glasswing. Partners identified at least 129,000 verified vulnerabilities between April and July 2026, while Anthropic's own initiatives found another 5,500 flaws by October.
More than 33,000 vulnerabilities were classified as high-severity or critical. The company says the numbers represent only part of the results because not all participants provided complete data.
Internal tests also indicated that reducing blocks expands the model's capability in authorized offensive tasks. In CyScenarioBench, Claude Opus 5.5 showed blocks in 46 of 50 attempts under Defense Access rules, while there were no blocks in the 50 tests conducted with Red Team Access.
The updated program will be available through Claude Platform, Google Cloud Vertex AI and Microsoft Foundry. On Amazon Bedrock, access will depend on eligibility for Anthropic's enterprise safeguards.
